roadmap
What is being built next, and how to get your idea onto the list
This is a plan, not a promise of dates. What it is, though, is honest: everything below is either already in progress or genuinely intended, and most of it arrived because somebody asked for it. The stops are in order; there are no dates on them.
Updated · Vitalii Yemets
Next stops, in order
Nemotron 3.5 as the default model
The streaming model becomes the one a fresh install starts with: fast, forty languages, words on the plate while you are still talking. Whisper stays in the catalogue for translation.
The first-run wizard, reworked
Fewer steps, a model chosen for your language without a question, a real dictation to try before the window closes.
Commands that edit text, not only insert it
Voice commands that change what is already at the cursor: delete the last sentence, make it a heading, turn it into a list. Today commands only insert.
Batch transcription of audio and video files
Drop a recording on the window and get the text out, with the same models, replacements and prompts as live dictation.
Updates that download only what changed
An update fetches the difference instead of the whole installer, so a new version costs a few megabytes, not thirty-five, and never touches the models.
The built-in help in every interface language
The full help inside the program exists in two languages; the other six still show the older, shorter text.
Microsoft Store and signed builds
A code-signing certificate, so Windows stops warning about an unknown publisher, then the program in the Microsoft Store: installed and updated from a place Windows already trusts.
Linux
The same program on Linux: the engines already run there, the tray, the shortcut hook and the plate do not yet.
Behind us
Push-to-talk dictation into any program
Hold two keys, speak, let go: the text lands where the cursor was, in any program with a text field.
Nine speech models, one per language
Whisper, Nemotron 3.5 with live text while you speak, Parakeet, Canary, GigaAM and more. The program keeps one per language.
Translation into seven languages
Speak in your language, get the text in English, German, Spanish, French, Italian, Polish or Ukrainian.
Post-processing by a local editor model
A small language model on your disk takes out fillers, fixes punctuation and changes tone. Prompts are yours to edit, or point it at any API.
Dictionary, replacements, voice commands
Rare words the model should lean towards, guaranteed replacements, and phrases that act instead of landing in the text.
History with exclusions
A list of what you dictated, text only, off by default, with programs you can leave out.
Seven designs, light and dark
Editor, Fluent, Studio, Soft, Document, Neon and Terminal, all set in IBM Plex Mono.
Installer and portable build
An installer that picks the folder and the first model, or a folder you unpack anywhere. Updates never re-download the models.
Eight interface languages
English, German, Spanish, French, Italian, Polish, Ukrainian and Russian, or it follows Windows.
First-run wizard
Five short steps on the first launch: interface language, dictation language and its model, shortcut and microphone, a field to try it in, start with Windows.
If you need a feature, ask — it gets built
Small things usually appear in the next release. Larger ones get a straight answer about when, or an explanation of why not. Nothing is ignored, and there is no voting queue to climb.
Open an issue
The best place for anything with steps, screenshots or a log. It stays public, so other people can add to it and see when it is done.
Request a featureReport a problem
Something broken, wrong or confusing. A version number and the last lines of the log are enough to start; the log is in English and never contains what you dictated.
Report a problemWrite directly
If GitHub is not for you, mail works just as well. Replies come from the person who writes the program, usually within a couple of days.
holdtotype@outlook.comWhat to put in a request
- What you were trying to do, in your own words.
- What the program does instead today.
- How often it happens — every dictation, or once a week.
- Your language and the model you use, if it might matter.
What happens then
- You get an answer, not an automated one.
- If it is small and clearly right, it lands in the next release.
- If it is large, it goes on this page so you can see where it stands.
- When it ships, the release notes say who asked for it, if you want them to.
Missing something? Say so.
The list above exists because people wrote in. Yours can be on it by tomorrow.